Nikon skin tones (part 2)

This is a continuation of the skin tone discussion thread.

Member said:

Let me simply interject here that you are conceding above that your comparisons involved two entirely separate RAW processing loops post WB normalization. Basically you are saying that you tried to make the D800 conversion look like the D700 using an entirely different profile in addition to manually tweaking contrast, whites, and saturation sliders.

Click to expand...

The RAW processing per camera/sensor is intrinsically different. When two development profiles have the same name, it doesn't mean the development is the same (this is also not claimed by the respective manufactures). All comparison experiments mentioned before, including your own experiment is a proof of that.

Especially the cross processing shown by fuzznut is telling a lot; the D700 picture starts to look like the D800 picture when the D800 profile is used.

Adapting one of these profiles to make their outcome look more alike is a perfect valid approach. It at least takes away the assertion there is an intrinsic green hue in the D800 RAW pictures. On top of that:

1. The RAWdigger histograms shows there is no significant difference in the source files.

2. On DXOmark you can see that the color variations per area are almost the same (because of the pixel pitch difference, and as shown by DXO mark, you need to compare for one ISO step difference), see the tab "Full CS" at DXOmark for the D700 and D800 .







3. Imatest results on imaging-resource show that the noise profile per color is the same under various lightness conditions.

Member said:

You are doing an exercise to prove that one can "fix" a D800 image to look like a D700 image within Lightroom/Photoshop.

Click to expand...

I'm not fixing an image, I fix the development flow by making them similar in terms of their output. With a bit more work I could probably take away the last differences so that the result of a D700 and D800 will be the same for whatever image you provide.

Member said:

Just realize, Marc, as best expressed by Stacey... that the debate here is not that global and/or local corrections can fix an image, it is that high density CMOS sensors have an inordinate amount of green shift baked into their images.

Click to expand...

You have failed to provide an experiment where we can firmly conclude this is the case. All experiments that you have been provided could easily be negated by a counter experiment.

E.g. your last experiment with NX2 change completely if you use Camera Neutral instead of Camera Standard. In that case the images look the same in terms of color shifts. Hence, we can (again...) rule out the source data to be the root cause, and point to the RAW development flow.

Member said:

So your entire position can be boiled down to this... the green shift phenomena does exist but it is solely relegated to Adobe Camera Standard color profile differentially transforming the color from the D700 versus the D800.

Click to expand...

Indeed. The cross processing shown by fuzznut is even confirming this from another perspective.

Member said:

I would conclude exactly the opposite. The chances are much higher that Adobe Camera Standard is giving consistent readout

Click to expand...

We shouldn't talk in terms of "chances", but in terms of repeatable and consistent experiments, that rule out all other causes before concluding a certain assertion is proven.

If you say the Adobe Standard profile shows greenish tints in D800 pictures, we definitely agree. I only claim that your assertion on green pixel noise has not been proven yet by your experiments, or comparison of pictures using a similar named RAW profile.

Member said:

and the difference we are seeing is the color noise variance between the D700's Nikon sensor and the D800's Sony sensor. This hypothesis is the only one that can sufficiently explain why the green shift is progressively worse as you move from 12 to 16 to 24 to 36 MP. It also explains why darker colors shift more than lighter colors in your comparison above... there's more color noise in darker shades than light.

Click to expand...

Again, look at the measurements done by DXOmark and imaging-resource, there is no single indication that your assertions are right.

Member said:

Observations on your comments complete, I'd like to challenge your hypothesis that green shift is solely due to the Adobe Standard profile. The same phenomena can be demonstrated entirely in Capture CNX2.

Click to expand...

It is not up to me to claim whether some specific noise exists. I only say with the proper development profiles, a D800 and D700 picture look similar in terms of their HSL profiles, so there is no reason to suspect noise from the experiments shown up to now.

Now some logic. The D800 has 36,3Mpix with a pitch of 23.72um2, hence an effective pixel coverage of 861 um. The D700 has 12.1Mi and a pitch, 70.9um2, hence 857,9um2. Not surprising that this is the same. This is also logical if you look at the math:



But it is not only the area, but also the depth of the metal layers in which the photons need to traverse to reach the light sensitive layer. As the D800 sensor is made in a denser CMOS technology (CMOS18) than the D700 sensor (CMOS35), this implies the smaller pixels can be made efficient. This probably attributes to the larger dynamic range for the D800.

Member said:

Even under CNX2 we see greenish skin tones and particularly so in shadow areas of the face like to the left of the woman's nose above.

Click to expand...

Redo your experiment with the Camera Neutral profile and the differences vanish. I conclude from that, that the Camera Standard profile causes differences. I did the experiment, and on the left side the result with Standard, and on the right Neutral for ISO6400 pictures are shown below (so therefore grainy). No reason to suspect green pixel noise when you cannot exclude the processing effect, and can show examples where it vanishes by taking another profile.



The experiment of fuzznut shows you can even reverse the conclusion by swapping the profiles. It will make the D700 look green, and a D800 look red-yellow. Hence, there is really no reason to suspect green pixel noise when you see green hues/shifts in a processed RAW file.

A side track, Adobe initially provided profiles where the D700 would suffer from magenta/green shifts in grey shadow areas. See the following thread. How about calling this noise, and claiming the new profiles just filter the noise? ;-)

Member said:

The conclusion here is simple and just as predicted. Regardless of the RAW convertor used, a high density CMOS sensor will inevitably shift colors towards green because it fundamentally requires more base amplification to drive its less sensitive pixels. The phenomena is both very real and easily reproduced by anyone.

Click to expand...

Three time the amount of amplifiers on 1/3rd pixel area implies... the same S/N!!!.

Please come up with an experiment where you exclude the effect of RAW convertor profiles. If you haven't excluded that effect, you have no backup for your conclusions on green pixel noise. Alternatively, come up with different measurements in line with DXOmark and imaging-resource that show green noise exists. Otherwise your statement on green pixel noise just an assertion or hypothesis.

I missed this post. Case closed!

fuzznut said:

I missed this post. Case closed!

Click to expand...

Quite often we have inadequate colour transforms in raw converters. That is the result of colour profiling tools (physical, like targets, and software) not up to the challenge of luminosity levels below L*=20 and above L*=90; D65 being theoretical light source; non-linearity in shadows (due to glare and flare) not being compensated leading to colour shifts.

Here is neutrality report for D800 at ISO 3200



And this one is for D700, same ISO 3200:



Overall, D800 looks a little better - down to L*=20. The problem with shadows is mainly due to the fact that profiling target itself does not go below L*=20, so the profiling software needs to guess what happens between 0 and 20 and extrapolate. In the presence of both noise and flare extrapolation starts to fail. It is also obvious from the D800 neutrality report that the noise affects mostly the shadows of red and blue channels (as fully expected), while green maintains decent linearity, contrary to some suggesting that noise in green is to be blamed.

But as for colour error, here is the report for D800 profiled at ISO 3200:



And this one is for D700, also ISO 3200:



Both show deltaE2000 error below just noticeable (JND is when deltaE reaches 2). So, no colour shift to speak of. When the shift occurs it is the artifact of colour transforms and colour profiles.

--

Thanks!

As expected, pretty much puts the whole green cast nonsense to bed...

-m

anotherMike said:

Thanks!

As expected, pretty much puts the whole green cast nonsense to bed...

-m

Click to expand...

Try saying that again when this thread gets shut down to post limits....

Marc Heijligers said:

This is a continuation of the skin tone discussion thread.

Click to expand...

Thanks for continuing the thread, Marc. Clearly there are a lot of people interested in the topic and wanting to drill further into it. I'm personally still a bit delayed in responding thanks to Father's Day falling on last weekend so this is going to be a placeholder post to say that I will need some more time to independently confirm the image processing that fuzznut and you have since posted on.

In the meantime here's a summary of where we agree and disagree:

1. We all agree there is visible green shift in D800 images rendered in Adobe Lightroom and Capture CNX2 at least when using Adobe Camera Standard or Nikon's Camera Standard profile, respectively. I think it's important to call this out specifically because prior to our work here there were outright denials by many D800 and D610 fanboys that green shift even existed. Now that we can all acknowledge that it is real, we can move to the discussion of what causes it.
2. I believe it is driven by elevated green channel noise due to the doubled ratio of green pixels in the RGGB Bayer CFA in conjunction with the higher sensor amplification required to drive the tiny, less light sensitive pixels in the D610 and D800.
3. You all believe it is driven solely by poorly written RAW color transforms in certain camera profiles and that substituting a different or alternate camera profile negates the green shift.

After I test the profiles in question I'll also come back to you in detail on the other technical points you added.

fPrime

Marc Heijligers said:

If you say the Adobe Standard profile shows greenish tints in D800 pictures, we definitely agree. I only claim that your assertion on green pixel noise has not been proven yet by your experiments, or comparison of pictures using a similar named RAW profile.

Click to expand...

Marc have you studied this enough to know if this (re: Adobe standard) is true in all cases, only in underexposed shots, shadows only, etc...? Have you compared adobe standard vs. camera neutral enough to be sure this is true? I don't mean to complicate things but in my opinion camera neutral gives pretty crappy skin tones. I played with 3 images in Lightroom, with my eyedropper on a skin tone I changed the profile a few times from adobe standard to camera neutral and while all three R-G-B numbers changed the ratios didn't change enough to conclude one was more "green".

Nexu1 said:

Marc Heijligers said:

If you say the Adobe Standard profile shows greenish tints in D800 pictures, we definitely agree. I only claim that your assertion on green pixel noise has not been proven yet by your experiments, or comparison of pictures using a similar named RAW profile.

Click to expand...

Marc have you studied this enough to know if this (re: Adobe standard) is true in all cases, only in underexposed shots, shadows only, etc...?

Click to expand...

It is impossible to elaborate towards "all" cases. One can deny the existence of gravity by claiming one could imagine there is an object that when loosened out of the hands, will disappear in the sky instead if falling down. These type of assertions are nice for philosophical exercises, but if you want to work on the stability of a car on a road, I guess it makes more sense to just assume gravity exists, and work with the scientific material we have at hand.

The assertion that the greenish tint is due to green pixel noise, based on looking at the outcome of a RAW development flow is what is my objection. You have to prove that the RAW development process is the same and transparent, and simple experiments and analysis show that this is not the case. Therefore, just observing the output of such a RAW development process will give intrinsic different output, and one cannot just point to certain properties of the input file being the single cause of that.

Nexu1 said:

Have you compared adobe standard vs. camera neutral enough to be sure this is true?

Click to expand...

I haven't compared it in great depth, and it was not my intent to show I've aligned the development settings such that they will always give exactly the same results (which I mentioned before). In order to make them truly the same, you need a lot of measurements under different type of color temperatures and light intensities; but this is not the purpose of my experiment.

I've only shown that with minimal adaptions, the so called green hue can be eliminated. This only shows that the assertion that green hue/noise is intrinsic to the sensor cannot be made by just comparing 2 different development flow from 2 different cameras.

Nexu1 said:

I don't mean to complicate things but in my opinion camera neutral gives pretty crappy skin tones. I played with 3 images in Lightroom, with my eyedropper on a skin tone I changed the profile a few times from adobe standard to camera neutral and while all three R-G-B numbers changed the ratios didn't change enough to conclude one was more "green".

Click to expand...

I agree that the Nikon D800 development flow tends towards a green hue, just like the D700 tends to a red hue, and my RX1 tends towards a blue hue. Obviously manufacturers choose for such a profile when releasing a camera.

The following article shows an interesting visualization of different color profiles for the Canon 5D MKII: http://chromasoft.blogspot.nl/2009/02/visualizing-dng-camera-profiles-part-1.html

ditch Lightroom asap. Camera Raw/Lightroom does an awful job of processing Nikon files.

Capture One Pro is sooooooooo much better and gives your photos much better skin tone and shadow detail. The learning curve really isn't too difficult if you know lightroom and you actually spend way less time making adjustments and trying to fix color

Even Aperture is better than Lightroom in terms image quality, but Apple has been neglecting Aperture and I think they might kill it off or just merge it with iPhoto, in which case it will be terrible.

I guess one man's meat is another's poison. I couldn't get Capture One off my computer fast enough the two times I tried it with the D800e. Same for Photo Ninja. Color was simply terrible. Lightroom/CC it will be for me for the foreseeable. Gorgeous!

I am very happy with Lightroom myself.

My understanding is that the Adobe Standard profile has hue twists. As a result colors can shift when exposure adjustments are applied in Lightroom. Adobe's reason for doing this is beyond my understanding. I'll leave that for others.

I'm not sure whether Camera Neutral is twisted or not.

Profiles generated with the QPcard software or ColorChecker Passport are not twisted.

Perhaps someone more knowledgeable than me can comment about whether twisted profiles could be what some are disliking about Lightroom 5.

I tried Capture One Pro 6 against Lightroom 4 sometime ago. IIRC, I found Lightroom to be superior in shadow recovery, and at least as good as C1P6 for color. C1P7 caught up with Lightroom 5 in some ways, but I didn't find anything about it to warrant changing from Lightroom 5.

Robin Casady said:

My understanding is that the Adobe Standard profile has hue twists. As a result colors can shift when exposure adjustments are applied in Lightroom. Adobe's reason for doing this is beyond my understanding. I'll leave that for others.

Click to expand...

Indeed.

With dcptool you are able to filter these hue twists out of the development files, and apply them to the respective pictures again. The results are shown underneath. These are the ISO6400 files, with the untwisted Adobe Standard profiles applied. Color noise filtering was put at 0, and only white balancing and exposure was aligned (to compensate for the dpreview differences).

As you can see, the results are merely indistinguishable.



This again shows; no reason to suspect green noise or hue shifts as a result from the sensor.

fPrime said:

1. I believe it is driven by elevated green channel noise due to the doubled ratio of green pixels in the RGGB Bayer CFA in conjunction with the higher sensor amplification required to drive the tiny, less light sensitive pixels in the D610 and D800.

Click to expand...

That does not make sense. The "doubled" ratio of green pixels is the same for the D700 and D800, all pixels are equally affected by the decrease in size, and the higher sensor amplification would be the same for all pixels.

The decrease in pixel size affects all pixels in the same way. The fact that there are 2 green pixels for 1 red and 1 blue does change that reality.

And I still don't see that green cast in skin tones in my pictures

TOF guy said:

fPrime said:

1. I believe it is driven by elevated green channel noise due to the doubled ratio of green pixels in the RGGB Bayer CFA in conjunction with the higher sensor amplification required to drive the tiny, less light sensitive pixels in the D610 and D800.

Click to expand...

That does not make sense. The "doubled" ratio of green pixels is the same for the D700 and D800, all pixels are equally affected by the decrease in size, and the higher sensor amplification would be the same for all pixels.

Click to expand...

Indeed.

TOF guy said:

And I still don't see that green cast in skin tones in my pictures

Click to expand...

What I see in practice is that the D800 white balance tends to hint at the green side instead of the purple side. And the D700 tends to saturate the reds more. I consider the D800 to be more accurate, where I have to do less post processing on skin tones than I used to do with my D700.

TOF guy said:

fPrime said:

1. I believe it is driven by elevated green channel noise due to the doubled ratio of green pixels in the RGGB Bayer CFA in conjunction with the higher sensor amplification required to drive the tiny, less light sensitive pixels in the D610 and D800.

Click to expand...

That does not make sense. The "doubled" ratio of green pixels is the same for the D700 and D800, all pixels are equally affected by the decrease in size, and the higher sensor amplification would be the same for all pixels.

Click to expand...

If anything, green channel has the least amount of noise, and having 2x times green pixels only helps in that regards.

"Higher sensor amplification", whatever it means, is yet another unsupported statement. I asked Mr. fPrime what is the basis for this statement, but of course he ignored my request. But again, green being most sensitive requires less boost (analogue or digital, through white balance).